* [PATCH v2 0/2] PM: runtime: add new runtime metrics and summary
@ 2026-11-16 9:36 Joshua Crofts
2026-11-16 9:36 ` [PATCH v2 1/2] PM: runtime: add additional suspend/resume metrics Joshua Crofts
2026-11-16 9:36 ` [PATCH v2 2/2] PM: runtime: add debugfs entry for pm_runtime_summary Joshua Crofts
0 siblings, 2 replies; 3+ messages in thread
From: Joshua Crofts @ 2026-11-16 9:36 UTC (permalink / raw)
To: Rafael J. Wysocki, Len Brown, Pavel Machek, Greg Kroah-Hartman,
Danilo Krummrich, Johan Hovold, Samuel Wu
Cc: linux-pm, driver-core, linux-kernel, Joshua Crofts
Add a new set additional PM runtime metrics to improve power management
debugging and reporting (these metrics may be useful to userspace utilities
such as Powertop).
MOTIVATION:
The current implementation only measures the duration a device spends in
suspended or active mode, which is very limiting. We can't find out how many
times a device changed state or what its shortest suspend cycle was.
This overview would be useful for detecting power "thrashing" and tuning
autosuspend delays. Currently, sysfs only shows the total time a device spent
in active and suspended mode; from an energy standpoint, a continuous 50
second suspension is different from a device cycling 1000 times for 50ms each
(these transitions can be pretty power-hungry). This could help userspace
utilities like PowerTOP to more effectively determine what is causing an
increased power consumption.
NEW FEATURES:
This series introduces the following new metrics:
- resume_count : number of successful transitions to active
- suspend_count: number of successful transitions to suspended
- last_suspended_time: duration of the most recent sleep cycle (ns)
- min_suspended_time / max_suspended_time: shortest and longest sleep
cycle durations to help detect power thrashing (ns)
By reading the debugfs file /sys/kernel/debug/pm_runtime_summary, you get
the following table describing the power stats of each device with
runtime PM enabled:
Device Status Resumes Suspends Active(ms) Suspend(ms) Last(ms) Min(ms) Max(ms)
--------------------------------------------------------------------------------------------------------------------------
0000:00:02.0 active 1 0 45688 0 0 0 0
0000:00:1f.0 active 1 0 45687 0 0 0 0
0000:00:1f.2 active 1 0 45686 0 0 0 0
0000:00:1f.3 suspended 1 1 1658 44028 0 0 0
00:03:0 active 15 14 19627 25899 42 5 14811
00:03:0.0 active 14 13 19626 25900 42 5 14811
TESTING:
Tested with an x86_64 defconfig kernel and QEMU.
AI ASSISTANCE:
Gemini was used to prepare this series.
Signed-off-by: Joshua Crofts <crofts@kernel.org>
---
Changes in v2:
- Drop sysfs patch as it would have a non-negligible impact on boot times and
the debugfs summary is enough (Samuel Wu)
- Expand cover letter
- Fix Sashiko findings - typo in commit message, race condition, duplicate
function call
- Link to v1: https://lore.kernel.org/r/20261110-pm-runtime-summary-v1-0-12ebf283db78@kernel.org
---
Joshua Crofts (2):
PM: runtime: add additional suspend/resume metrics
PM: runtime: add debugfs entry for pm_runtime_summary
drivers/base/power/runtime.c | 162 +++++++++++++++++++++++++++++++++++++++++--
include/linux/pm.h | 12 ++++
2 files changed, 169 insertions(+), 5 deletions(-)
---
base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4
change-id: 20261110-pm-runtime-summary-8dd29a1383bc
Best regards,
--
Joshua Crofts <crofts@kernel.org>
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH v2 1/2] PM: runtime: add additional suspend/resume metrics
2026-11-16 9:36 [PATCH v2 0/2] PM: runtime: add new runtime metrics and summary Joshua Crofts
@ 2026-11-16 9:36 ` Joshua Crofts
2026-11-16 9:36 ` [PATCH v2 2/2] PM: runtime: add debugfs entry for pm_runtime_summary Joshua Crofts
1 sibling, 0 replies; 3+ messages in thread
From: Joshua Crofts @ 2026-11-16 9:36 UTC (permalink / raw)
To: Rafael J. Wysocki, Len Brown, Pavel Machek, Greg Kroah-Hartman,
Danilo Krummrich, Johan Hovold, Samuel Wu
Cc: linux-pm, driver-core, linux-kernel, Joshua Crofts
The current implementation only logs the total accumulated time a device
spends in suspended or active states.
Add tracking for:
- resume_count : number of successful transitions to active
- suspend_count: number of successful transitions to suspended
- last_suspended_time: duration of the most recent sleep cycle (ns)
- min_suspended_time / max_suspended_time: shortest and longest sleep
cycle durations to help detect power thrashing (ns)
Signed-off-by: Joshua Crofts <crofts@kernel.org>
---
drivers/base/power/runtime.c | 76 +++++++++++++++++++++++++++++++++++++++++---
include/linux/pm.h | 12 +++++++
2 files changed, 83 insertions(+), 5 deletions(-)
diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
index 304a2b12342082ffacce32aa9c2e863466b0d4ae..3085318e9c32fce892def5a901aae300f473ea2d 100644
--- a/drivers/base/power/runtime.c
+++ b/drivers/base/power/runtime.c
@@ -5,6 +5,8 @@
* Copyright (c) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
* Copyright (C) 2010 Alan Stern <stern@rowland.harvard.edu>
*/
+#include <linux/pm.h>
+#include <linux/timekeeping.h>
#include <linux/sched/mm.h>
#include <linux/ktime.h>
#include <linux/hrtimer.h>
@@ -103,11 +105,58 @@ static void update_pm_runtime_accounting(struct device *dev)
dev->power.active_time += delta;
}
+static void rpm_account_suspend(struct device *dev)
+{
+ dev->power.suspend_time = dev->power.accounting_timestamp;
+ dev->power.suspend_count++;
+}
+
+static void rpm_account_sleep_end(struct device *dev)
+{
+ u64 duration;
+
+ if (!dev->power.suspend_time)
+ return;
+
+ if (dev->power.accounting_timestamp <= dev->power.suspend_time)
+ return;
+
+ duration = dev->power.accounting_timestamp - dev->power.suspend_time;
+ dev->power.last_suspended_time = duration;
+
+ /* Update min/max extrema at the exact same moment as last_suspended_time */
+ if (!dev->power.min_suspended_time || duration < dev->power.min_suspended_time)
+ dev->power.min_suspended_time = duration;
+
+ if (duration > dev->power.max_suspended_time)
+ dev->power.max_suspended_time = duration;
+}
+
+static void rpm_account_resume_complete(struct device *dev)
+{
+ dev->power.resume_count++;
+ dev->power.suspend_time = 0;
+}
+
static void __update_runtime_status(struct device *dev, enum rpm_status status)
{
+ enum rpm_status old_status = dev->power.runtime_status;
+
update_pm_runtime_accounting(dev);
trace_rpm_status(dev, status);
dev->power.runtime_status = status;
+
+ if ((old_status == RPM_SUSPENDING || old_status == RPM_ACTIVE) &&
+ status == RPM_SUSPENDED)
+ rpm_account_suspend(dev);
+ else if (old_status == RPM_RESUMING && status == RPM_SUSPENDED)
+ dev->power.suspend_time = dev->power.accounting_timestamp;
+ else if (old_status == RPM_SUSPENDED && status != RPM_SUSPENDED)
+ rpm_account_sleep_end(dev);
+
+ if ((old_status == RPM_RESUMING || old_status == RPM_SUSPENDED) &&
+ status == RPM_ACTIVE)
+ rpm_account_resume_complete(dev);
}
static u64 rpm_get_accounted_time(struct device *dev, bool suspended)
@@ -1571,6 +1620,7 @@ EXPORT_SYMBOL_GPL(__pm_runtime_disable);
void pm_runtime_enable(struct device *dev)
{
unsigned long flags;
+ u64 now;
spin_lock_irqsave(&dev->power.lock, flags);
@@ -1587,12 +1637,21 @@ void pm_runtime_enable(struct device *dev)
dump_stack();
}
dev->power.last_status = RPM_INVALID;
- dev->power.accounting_timestamp = ktime_get_mono_fast_ns();
+ now = ktime_get_mono_fast_ns();
- if (dev->power.runtime_status == RPM_SUSPENDED &&
- !dev->power.ignore_children &&
- atomic_read(&dev->power.child_count) > 0)
- dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
+ if (dev->power.runtime_status == RPM_SUSPENDED) {
+ if (dev->power.suspend_time && dev->power.accounting_timestamp
+ && now > dev->power.accounting_timestamp)
+ dev->power.suspend_time += now - dev->power.accounting_timestamp;
+ else
+ dev->power.suspend_time = now;
+
+ if (!dev->power.ignore_children &&
+ atomic_read(&dev->power.child_count) > 0)
+ dev_warn(dev, "Enabling runtime PM for inactive device with active children\n");
+ }
+
+ dev->power.accounting_timestamp = now;
out:
spin_unlock_irqrestore(&dev->power.lock, flags);
@@ -1861,6 +1920,13 @@ void pm_runtime_init(struct device *dev)
pm_suspend_ignore_children(dev, false);
dev->power.runtime_auto = true;
+ dev->power.resume_count = 0;
+ dev->power.suspend_count = 0;
+ dev->power.suspend_time = 0;
+ dev->power.last_suspended_time = 0;
+ dev->power.min_suspended_time = 0;
+ dev->power.max_suspended_time = 0;
+
dev->power.request_pending = false;
dev->power.request = RPM_REQ_NONE;
dev->power.deferred_resume = false;
diff --git a/include/linux/pm.h b/include/linux/pm.h
index ef3f1310e749c34d4aff75a2fd7e95b9f2ef3a10..7cd1350ea327be4e800175c2fc0e3b3e35c3ea57 100644
--- a/include/linux/pm.h
+++ b/include/linux/pm.h
@@ -747,6 +747,12 @@ struct pm_subsys_data {
* state.
* @accounting_timestamp: Timestamp in nanoseconds of the last runtime PM state
* accounting update.
+ * @resume_count: Number of transitions from suspended -> active
+ * @suspend_count: Number of transitions from active -> suspended
+ * @suspend_time: Timestamp when device entered RPM_SUSPENDED
+ * @last_suspended_time: Residency of the last sleep period (ns)
+ * @min_suspended_time: Shortest sleep duration
+ * @max_suspended_time: Longest sleep duration
* @subsys_data: Subsystem-specific power management data.
* @set_latency_tolerance: Callback for setting latency tolerance.
* @qos: Per-device PM Quality of Service (QoS) constraints.
@@ -816,6 +822,12 @@ struct dev_pm_info {
u64 active_time;
u64 suspended_time;
u64 accounting_timestamp;
+ u64 resume_count;
+ u64 suspend_count;
+ u64 suspend_time;
+ u64 last_suspended_time;
+ u64 min_suspended_time;
+ u64 max_suspended_time;
#endif
struct pm_subsys_data *subsys_data; /* Owned by the subsystem. */
void (*set_latency_tolerance)(struct device *, s32);
--
2.47.3
^ permalink raw reply related [flat|nested] 3+ messages in thread
* [PATCH v2 2/2] PM: runtime: add debugfs entry for pm_runtime_summary
2026-11-16 9:36 [PATCH v2 0/2] PM: runtime: add new runtime metrics and summary Joshua Crofts
2026-11-16 9:36 ` [PATCH v2 1/2] PM: runtime: add additional suspend/resume metrics Joshua Crofts
@ 2026-11-16 9:36 ` Joshua Crofts
1 sibling, 0 replies; 3+ messages in thread
From: Joshua Crofts @ 2026-11-16 9:36 UTC (permalink / raw)
To: Rafael J. Wysocki, Len Brown, Pavel Machek, Greg Kroah-Hartman,
Danilo Krummrich, Johan Hovold, Samuel Wu
Cc: linux-pm, driver-core, linux-kernel, Joshua Crofts
Add a debugfs entry for showing the runtime statistics for devices with
PM runtime enabled.
Signed-off-by: Joshua Crofts <crofts@kernel.org>
---
drivers/base/power/runtime.c | 86 ++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 86 insertions(+)
diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
index 3085318e9c32fce892def5a901aae300f473ea2d..2012c49882092fdfbe57b8d8d6e997328de207cf 100644
--- a/drivers/base/power/runtime.c
+++ b/drivers/base/power/runtime.c
@@ -14,6 +14,8 @@
#include <linux/pm_runtime.h>
#include <linux/pm_wakeirq.h>
#include <linux/rculist.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
#include <trace/events/rpm.h>
#include "../base.h"
@@ -2197,3 +2199,87 @@ bool pm_runtime_need_not_resume(struct device *dev)
}
#endif /* CONFIG_PM_SLEEP */
+
+#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_PM_SLEEP)
+static void rpm_summary_show_dev(struct device *dev, void *data)
+{
+ struct seq_file *s = data;
+ unsigned long flags;
+ u64 active_time, suspended_time;
+ u64 last_sleep, min_sleep, max_sleep;
+ u64 resumes, suspends;
+ const char *status_str;
+ bool enabled;
+
+ spin_lock_irqsave(&dev->power.lock, flags);
+
+ enabled = pm_runtime_enabled(dev);
+ if (!enabled && !dev->power.resume_count && !dev->power.suspend_count) {
+ spin_unlock_irqrestore(&dev->power.lock, flags);
+ return;
+ }
+
+ update_pm_runtime_accounting(dev);
+
+ switch (dev->power.runtime_status) {
+ case RPM_ACTIVE:
+ status_str = "active";
+ break;
+ case RPM_SUSPENDED:
+ status_str = "suspended";
+ break;
+ case RPM_SUSPENDING:
+ status_str = "suspending";
+ break;
+ case RPM_RESUMING:
+ status_str = "resuming";
+ break;
+ default:
+ status_str = "unknown";
+ break;
+ }
+
+ active_time = dev->power.active_time;
+ suspended_time = dev->power.suspended_time;
+ resumes = dev->power.resume_count;
+ suspends = dev->power.suspend_count;
+ last_sleep = dev->power.last_suspended_time;
+ min_sleep = dev->power.min_suspended_time;
+ max_sleep = dev->power.max_suspended_time;
+
+ spin_unlock_irqrestore(&dev->power.lock, flags);
+
+ do_div(active_time, NSEC_PER_MSEC);
+ do_div(suspended_time, NSEC_PER_MSEC);
+ do_div(last_sleep, NSEC_PER_MSEC);
+ do_div(min_sleep, NSEC_PER_MSEC);
+ do_div(max_sleep, NSEC_PER_MSEC);
+
+ seq_printf(s, "%-32s %-12s %8llu %8llu %12llu %12llu %10llu %10llu %10llu\n",
+ dev_name(dev), enabled ? status_str : "disabled",
+ resumes, suspends, active_time, suspended_time,
+ last_sleep, min_sleep, max_sleep);
+}
+
+static int pm_runtime_summary_show(struct seq_file *s, void *unused)
+{
+ seq_printf(s, "%-32s %-12s %8s %8s %12s %12s %10s %10s %10s\n",
+ "Device", "Status", "Resumes", "Suspends",
+ "Active(ms)", "Suspend(ms)", "Last(ms)", "Min(ms)", "Max(ms)");
+ seq_puts(s, "--------------------------------------------------------------------------------------------------------------------------\n");
+
+ dpm_for_each_dev(s, rpm_summary_show_dev);
+
+ return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(pm_runtime_summary);
+
+static int __init pm_runtime_debugfs_init(void)
+{
+ debugfs_create_file("pm_runtime_summary", 0444, NULL, NULL,
+ &pm_runtime_summary_fops);
+
+ return 0;
+}
+late_initcall(pm_runtime_debugfs_init);
+#endif /* defined(CONFIG_DEBUG_FS) && defined(CONFIG_PM_SLEEP) */
--
2.47.3
^ permalink raw reply related [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-11-16 9:36 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 9:36 [PATCH v2 0/2] PM: runtime: add new runtime metrics and summary Joshua Crofts
2026-11-16 9:36 ` [PATCH v2 1/2] PM: runtime: add additional suspend/resume metrics Joshua Crofts
2026-11-16 9:36 ` [PATCH v2 2/2] PM: runtime: add debugfs entry for pm_runtime_summary Joshua Crofts
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox